Artificial Intelligence (AI) Models for Cardiovascular Disease Risk Prediction in Primary and Ambulatory Care: A Scoping Review

Artificial Intelligence (AI) Models for Cardiovascular Disease Risk Prediction in Primary and Ambulatory Care: A Scoping Review | medRxiv /* */ /* */ <!-- <!-- /*! * yepnope1.5.4 * (c) WTFPL, GPLv2 */ (function(a,b,c){function d(a){return"[object Function]"==o.call(a)}function e(a){return"string"==typeof a}function f(){}function g(a){return!a||"loaded"==a||"complete"==a||"uninitialized"==a}function h(){var a=p.shift();q=1,a?a.t?m(function(){("c"==a.t?B.injectCss:B.injectJs)(a.s,0,a.a,a.x,a.e,1)},0):(a(),h()):q=0}function i(a,c,d,e,f,i,j){function k(b){if(!o&&g(l.readyState)&&(u.r=o=1,!q&&h(),l.onload=l.onreadystatechange=null,b)){"img"!=a&&m(function(){t.removeChild(l)},50);for(var d in y[c])y[c].hasOwnProperty(d)&&y[c][d].onload()}}var j=j||B.errorTimeout,l=b.createElement(a),o=0,r=0,u={t:d,s:c,e:f,a:i,x:j};1===y[c]&&(r=1,y[c]=[]),"object"==a?l.data=c:(l.src=c,l.type=a),l.width=l.height="0",l.onerror=l.onload=l.onreadystatechange=function(){k.call(this,r)},p.splice(e,0,u),"img"!=a&&(r||2===y[c]?(t.insertBefore(l,s?null:n),m(k,j)):y[c].push(l))}function j(a,b,c,d,f){return q=0,b=b||"j",e(a)?i("c"==b?v:u,a,b,this.i++,c,d,f):(p.splice(this.i++,0,a),1==p.length&&h()),this}function k(){var a=B;return a.loader={load:j,i:0},a}var l=b.documentElement,m=a.setTimeout,n=b.getElementsByTagName("script")[0],o={}.toString,p=[],q=0,r="MozAppearance"in l.style,s=r&&!!b.createRange().compareNode,t=s?l:n.parentNode,l=a.opera&&"[object Opera]"==o.call(a.opera),l=!!b.attachEvent&&!l,u=r?"object":l?"script":"img",v=l?"script":u,w=Array.isArray||function(a){return"[object Array]"==o.call(a)},x=[],y={},z={timeout:function(a,b){return b.length&&(a.timeout=b[0]),a}},A,B;B=function(a){function b(a){var a=a.split("!"),b=x.length,c=a.pop(),d=a.length,c={url:c,origUrl:c,prefixes:a},e,f,g;for(f=0;f<d;f++)g=a[f].split("="),(e=z[g.shift()])&&(c=e(c,g));for(f=0;f<b;f++)c=x[f](c);return c}function g(a,e,f,g,h){var i=b(a),j=i.autoCallback;i.url.split(".").pop().split("?").shift(),i.bypass||(e&&(e=d(e)?e:e[a]||e[g]||e[a.split("/").pop().split("?")[0]]),i.instead?i.instead(a,e,f,g,h):(y[i.url]?i.noexec=!0:y[i.url]=1,f.load(i.url,i.forceCSS||!i.forceJS&&"css"==i.url.split(".").pop().split("?").shift()?"c":c,i.noexec,i.attrs,i.timeout),(d(e)||d(j))&&f.load(function(){k(),e&&e(i.origUrl,h,g),j&&j(i.origUrl,h,g),y[i.url]=2})))}function h(a,b){function c(a,c){if(a){if(e(a))c||(j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}),g(a,j,b,0,h);else if(Object(a)===a)for(n in m=function(){var b=0,c;for(c in a)a.hasOwnProperty(c)&&b++;return b}(),a)a.hasOwnProperty(n)&&(!c&&!--m&&(d(j)?j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}:j[n]=function(a){return function(){var b=[].slice.call(arguments);a&&a.apply(this,b),l()}}(k[n])),g(a[n],j,b,n,h))}else!c&&l()}var h=!!a.test,i=a.load||a.both,j=a.callback||f,k=j,l=a.complete||f,m,n;c(h?a.yep:a.nope,!!i),i&&c(i)}var i,j,l=this.yepnope.loader;if(e(a))g(a,0,l,0);else if(w(a))for(i=0;i (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0];var j=d.createElement(s);var dl=l!='dataLayer'?'&l='+l:'';j.src='//www.googletagmanager.com/gtm.js?id='+i+dl;j.type='text/javascript';j.async=true;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-P4HH5NV'); Skip to main content Home About Submit ALERTS / RSS Search for this keyword Advanced Search Artificial Intelligence (AI) Models for Cardiovascular Disease Risk Prediction in Primary and Ambulatory Care: A Scoping Review Caoimhe Provost , John Broughan , Geoff McCombe , Mark O Kelly , Mark Ledwidge , Walter Cullen , Joseph Gallagher doi: https://doi.org/10.1101/2025.03.21.25324379 Caoimhe Provost 1 School of Medicine, University College Dublin , Belfield, Dublin, Ireland Find this author on Google Scholar Find this author on PubMed Search for this author on this site John Broughan 2 UCD School of Medicine, Clinical Research Centre St Vincent’s University Hospital , Dublin, Ireland Find this author on Google Scholar Find this author on PubMed Search for this author on this site For correspondence: jgallagher{at}ucd.ie Geoff McCombe 1 School of Medicine, University College Dublin , Belfield, Dublin, Ireland Find this author on Google Scholar Find this author on PubMed Search for this author on this site Mark O Kelly 1 School of Medicine, University College Dublin , Belfield, Dublin, Ireland Find this author on Google Scholar Find this author on PubMed Search for this author on this site Mark Ledwidge 1 School of Medicine, University College Dublin , Belfield, Dublin, Ireland Find this author on Google Scholar Find this author on PubMed Search for this author on this site Walter Cullen 1 School of Medicine, University College Dublin , Belfield, Dublin, Ireland Find this author on Google Scholar Find this author on PubMed Search for this author on this site Joseph Gallagher 1 School of Medicine, University College Dublin , Belfield, Dublin, Ireland Find this author on Google Scholar Find this author on PubMed Search for this author on this site Abstract Full Text Info/History Metrics Data/Code Preview PDF Abstract Background Mortality from cardiovascular disease (CVD) has seen a dramatic increase over the past decades, which has led to a significant increase in the development of risk prediction models. AI-based models have been proposed as a method of enhancing traditional risk models. This review aims to describe the present state of AI risk prediction models for cardiovascular disease in primary and ambulatory care research, and in particular to determine: the stage of development these models have reached, the AI approaches used, and identifying possible sources of bias or limitations in the AI models. Methods Using the Arksey and O’Malley scoping review method, this review searched Pubmed, EBSCOHost, and Web of Science databases between 2019 and 2024, and relevant studies were identified. Data extraction was performed on eligible included studies. Results 22,860 studies were screened, and 25 articles were identified. There was a lack of external validation (20% of models) and lack of clinical impact studies (0%) in this review. A variety of AI techniques were used. Both data and algorithmic biases were commonly identified. There was a lack of geographic variation in datasets (60% were based in the USA) and only 32% of studies reported race and ethnicity data. There was poor predictor and outcome standardization. Calibration was only reported in 24% of models. Conclusion Findings from this review highlight the lack of clinical impact studies and risk of bias in current AI based models. It provides evidence for future refinement and development of AI risk prediction models in cardiovascular disease. Author summary Cardiovascular disease (CVD) is a leading cause of mortality worldwide. In this paper, we undertake a scoping review of artificial intelligence (AI) models for predicting CVD risk in primary care (PC) and ambulatory care (AC) settings. Our findings underscore the critical role of AI in improving the prediction and prevention of cardiovascular diseases, while also highlighting the existing gaps and challenges in the application and validation of these models in primary and ambulatory care. We identified major biases in data from which models were developed, such as geographic discrepancies and inconsistencies among CVD predictors. Additionally, biases in the algorithm were significant issues that need to be addressed in the future. We emphasize the need for multi-centered, standardized, and externally validated models to ensure their utility and reliability in clinical settings. Our research underscores the importance of addressing both data and algorithmic biases, and the necessity of external validation and clinical trials to enhance the accuracy and clinical utility of these AI models in real-world settings. Background Mortality from cardiovascular disease (CVD) has seen a dramatic increase over the past few decades, rising from 12.1 million to 18.6 million cases worldwide from 1999 to 2019. 1 In the EU, CVD is the cause of 42% of deaths, which costs the EU approximately 192 million euros per year as per 2008. 2 Risk prediction models are used by clinicians to identify those most at risk of developing cardiovascular disease and to decide who may require interventions such as medications. Risk prediction models Risk prediction models are defined as “statistical models that combine information from several markers”. 3 (p.6) Risk prediction models produce a prediction of a patient’s risk based on the information inputted into the model. 3 A model and its output is only reliable and clinically useful as the quality or quantity of markers (also known as risk factors) put into the model. Common uses of risk prediction models prediction of sepsis 4 , acute kidney injury, 5 or CVD, such as coronary artery disease or heart failure. 6 However, there are some key issues that are associated with these models, such as overly optimistic performance metrics (which may lead to models that do not generalise well on the introduction of new data) 7 or the proper communication of the risk reliability and variability produced by the clinical prediction models. 8 Models may not perform as well in certain populations. For example age is typically a factor that is weighed heavily, which may cause younger patients may be more likely to receive an inaccurate or diminished risk prediction value. 6 The process of the development of a risk prediction model consists of nine main stages as outlined by Hassan et al 9 . Steps one and two consist of generating an objective for the model, as well as defining the parameters and risk factors included. Step three is when the researchers chose a data set in which to derive their model from. This may be from a singular clinic, a series of several clinics, or a national/international database with data on the risk factors intended to be used. Step four consists of the derivation, or the creation of the model, which can use deep learning and/or machine learning techniques, and should be collaborated on with a programming specialist 9 (to ensure AI governance, which refers to the legality and ethicality of AI in our lives 10 ). Step five concerns the validation of the model, both from the dataset it was derived from (internal validation), and datasets external to the set the model was derived from (external validation). The sixth step regards the interpretation of the model, or presentation of the outcome processed by the AI model. Any underlying assumptions of the model should be identified and addressed, such as bias. Steps seven and eight regard the licensing of the model as well as the maintenance, such as updating models with new information available (i.e. newly identified risk factors or populations). The final step involved the clinical assessment of the risk prediction models. This may include “the accuracy of the model predictions, physician and patient understanding and use of these probabilities, expected effectiveness of subsequent actions or interventions, and adherence to these”. 9 (p. 5) Complete adherence and completion of this nine stage process is essential to the proper development of a risk prediction model, as to ensure the smallest margin for error in the model programming and use. Artificial Intelligence in Healthcare AI has the potential to develop new and more accurate models. Artificial intelligence is defined as a machine-based system that is designed to operate with varying levels of autonomy and that may exhibit adaptiveness after deployment, and that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments. 8 These modern risk prediction models have been shown to play a significant role in healthcare and risk prevention as a whole. 11 These models have also shown that AI based models enhance clinical prediction in several domains, such as prognosis, risk assessment, and mortality prediction, compared to traditional statistics models. 12 Cardiovascular Disease There are many different risk factors for CVD, and they can differ significantly between different AI risk prediction models. AI models now are encompassing the evaluation of more wide-reaching and diverse set of risk factors. These may include data collected from wearable devices (despite their inferior performance compared to other data collection forms) 13 or CVD risk in conjunction with other disease diagnoses such as cancer or diabetes. 14 , 15 However, it is important to mention that AI models can be imperfect, as issues may arise such as bias (or systematic errors, such as in model choice, data used to train the model, or model evaluation, present in model predictions 16 ), accountability (“a set of protocols to evaluate the conduct of an entity, as well as an obligation to report or justify one’s actions, specially if these may result in any wrongdoing” 17 (p. 3) ), and transparency (ensuing information about the derivation and development of the model is accessible to the public 18 ). The presence or lack of these issues can lead either to a more or less accurate AI prediction model, and these issues may have a real-life impact on patient treatment and outcome. Various CVD AI risk prediction models do exist currently in the PC and AC setting. These models use various different tests routinely available in GP or ambulatory care settings with the goal of predicting cardiovascular disease or events or death. For example, results from routine blood tests, such as total cholesterol levels of serum creatinine levels, 19 as well as electrocardiogram results 20 may be used in order to derive these AI models. Gap in the Literature In current literature, there exists a gap in the knowledge consolidated and reviewing AI risk prediction models evaluating CVD in primary care (PC) and ambulatory care (AC). Cai et al. 2021 reviewed CVD AI risk prediction models and their validation, concluding that there was larger focus on developing new models than validating (or evaluating model performance/accuracy) existing ones 21 , but there is a notable lack of a focus on PC or AC accessible models. Silva et al. reported that errors in the AI models such as explainability, bias or data leakage need to be properly monitored and heavily analysed to assure both ethical and legal AI usage 22 , which has not been analysed in pre-existing PC or AC literature. Therefore, this paper aims to address three main aims: To determine what stage of development these models have reached. To determine what AI approaches have been used in the derivation and validation of these models. To identify possible sources of bias or limitations in the AI models identified. Methods This study chose to conduct this review using a scoping review methodology, as it aligns with the objective of gaining a complete and overarching review of the clinical value risk prediction models involving AI for CVD for primary care. Scoping reviews, which are defined as “knowledge syntheses used to collect, evaluate, and present findings from existing research on a topic” 23 (para. 5) are fundamental to the furthering and summarising existing knowledge from around literature and establishing gaps. This study followed the six-stage PRISMA scoping review method outlined by Arksey and O’Malley’s 24 , as well as Levac et al. 25 Stage 1: Identifying the research question With regard to both the gap in the previous knowledge surrounding CVD AI risk prediction models used in primary/ambulatory care settings and the objectives of this study, the research question was developed: What does the pre-existing literature report on artificial intelligence cardiovascular disease risk prediction in PC and AC, specifically: what stage of development these models have reached, what AI approaches have been used in the derivation and validation of these models, and what possible sources of bias or limitations in the AI models identified? Stage 2: Identifying relevant studies An initial search of databases including PubMed/Medline, Web of Science, and EBSCOHost was taken on 01/06/2024. To search these databases, a series of medical subject heading (MeSH) terms were identified in alignment with this study’s research question. These terms were then grouped together based on their subject matter, and so that resulting searches would include at least one search term of the four groups included (see appendix 1 ). Stage 3: Selecting studies Following the consolidation of the identified studies from the databases, duplicates were removed. Two reviewers read the identified studies titles and abstracts, and full texts were retrieved for all the studies. Studies found to be relevant to the study had a full-text review conducted by one researcher. This process is outlined in Figure 1 below Download figure Open in new tab Fig 1. PRISMA flow chart of this scoping review methodology Eligibility criteria were developed for this scoping review. Included studies were published between 2021 and 2024 and involved adult human participants. These studies also must include 1) a multivariable AI model relating to the risk prediction of CVD, 2) an outcome measure of either CVD or cardiovascular events or death and 3) conducted in a hospital outpatient or primary care setting. Excluded studies for this review included non-English papers, those studying paediatric patients only, as well as 1) studies based on single predictors only, 2) clinical prediction rules developed from patients with acute cardiovascular events or cardiac surgery only or hospital inpatients only, and 3) models involving tools not available routinely in primary care (see appendix 2 ). Stage 4: Charting the data After analysing all 25 full texts, a data extraction table was created. This included categories such as: Author and year of publication, location of study and datasets used, outcome and output of the risk prediction model, results of the AI risk prediction model incusing mentions of discrimination or calibration, the type of AI used, predictors used model development, limitations of the study, and the models most recent stage of validation. A copy of the table created is shown in Table 1 . View this table: View inline View popup Download powerpoint Table 1: The worldwide distribution of studies included in this review Stage 5: Collating, summarising, and reporting results This review focused primarily on emphasising key elements of the research found during stage 4 of the PRISMA methodology. In accordance with Tripod AI guidelines 26 as well as consultation from experts in the field of AI risk prediction models and primary/ambulatory care, six overarching result categories were described. Stage 6: Consultation This paper underwent a consultation exercise by two individuals in order to enhance its usefulness to policymakers, practitioners and other readers. 24 Upon this consultation, no further studies which fit the inclusion criteria were brought to the attention of the authors. However, one individual noted that there was a lack of the genetic and biomarker data such as natriuretic peptide data used in datasets to train the AI model, despite its growing popularity as a method of risk prediction for CVD. This clinical data would provide further information into the genetic predispositions, including epigenetic factors which may increase, decrease, or have a neutral effect on the likelihood of a patient to develop CVD. 27 It was also highlighted regarding the need to ensure adequate supervision of the models in clinical settings. Supervision in machine learning refers to the process of training a model using a labelled dataset, where the input data is paired with corresponding outputs (labels). This process allows the model to learn the relationship between inputs and outputs, enabling it to make predictions or decisions when presented with new, unseen data. This is important in clinical practice to ensure appropriate input data is utilised. 28 Results Study population and statistics In total, 25 full text articles were included in this review, all which were published between 2021 and 2024. These studies contained a combined 32 unique datasets from four distinct regions (North America, South America, Asia, and Europe), as shown in Figure 2 and Table 1 below. Download figure Open in new tab Fig 2. The worldwide distribution of datasets included in this review In the 25 studies included, the average age of participants was approximately 62.7 years old (41.2 29 -78.8 years 30 ). 52.1% of included participants were male, and of the 23 studies that did specify the sex ratios, 16 contained a cohort with a roughly balanced proportion of male to female participants (between 40% - 60% male). Eight studies included race and/or ethnicity statistics of their cohort. 31 , 32 , 33 , 34 , 35 , 36 , 37 , 38 59.8% of participants were White or European American, with the African American/Black/Black British population making up about 23.0% of all cohorts. While 21 included studies focused generally on the current population, five studies concerned sub-groups of the population, rather than the general public. These concerned oncology patients (n=1, 4.8%), 39 women and women’s health (n=1, 4.8%), 37 older patients (n=2, 9.6%), 30 , 40 prediabetes or diabetes outpatients (n=1, 4.8%), 40 and African-American patients (n=1, 4.8%). 41 The average sample size of all the studies was approximately 89783 participants (289 42 −691645 participants 43 ). The distribution of study sizes is shown in Figure 3 . A total of four different study settings were identified, which included an in person primary care setting, hospital (inpatient or outpatient) setting, online or telehealth surveys, and referral centre setting as shown in Figure 4 . Download figure Open in new tab Fig 3. Patient sample size for included studies Download figure Open in new tab Fig 4. Setting of data collection for included studies Setting of data collection A significant majority of the included texts (n=15) were conducted in a community or clinical care setting, followed by those in a hospital or inpatient setting (that used tools available to PC or AC practitioners) (n=6), 30 , 32 , 37 , 44 , 45 , 46 those in a virtual or online survey (n=2), 42 , 47 and studies conducted in a referral centre (n=1). 48 Of the 25 full-texts included, 11 specified that only completed data (i.e. patients with complete electronic health records and/or electrocardiograms) were included in the training or validation sets for the AI model. The minority of included studies used multiple databases (n=8) for the training (and the validation) of their risk prediction model 29 , 36 , 37 , 38 , 45 , 46 , 48 , 49 with the rest relying on one database in order to train, test, and validate the models Use of CVD predictors Across all models, 255 unique parameters used by the included AI risk prediction CVD models were identified. Age (n=18), body mass index (BMI) (n=13), and history of or current habit of smoking (n=10) were the most common parameters used in the models. Of the 255 parameters, 57 were found to have been included in models from two or more studies, and 27 in three or more papers, which are shown in Figure 5 . It is also important to highlight the presence of vague predictors, such as “12-lead raw electrocardiogram (ECG) data”, 50 but not the elements they extracted from that specific data. This may be an example of unsupervised training. Download figure Open in new tab Fig 5. The most used CVD predictors present in three or more included AI risk prediction models CVD outcome measurement and method All included texts in this review stated what their AI model was predicting by the patient outcome. The predictions of an unspecified CVD diagnosis were found to be the most common (n=6). 29 , 39 , 41 , 43 , 44 , 47 Other predictions included the onset or diagnosis of atrial fibrillation (n=5), 30 , 33 , 35 , 42 , 46 the development of heart failure (n=4), 31 , 36 , 40 , 47 prediction of mortality (with all but one describing predictions of all-cause mortality) (n=4), 32 , 44 , 45 , 48 “net adverse clinical events” such as all-cause death, myocardial infarction, stroke or major bleeding (n=3), 51 , 52 , 53 coronary artery disease events (n=2), 34 , 38 and aortic regurgitation development (n=1). 54 One of the models had a composite endpoint -prediction of mortality and/or cardiovascular events. 44 15 models mentioned periods of time their prediction was for, with the most common being exactly 1-year following the initial data collection (n=7), 32 , 34 , 35 , 43 , 48 , 52 , 54 as shown in Figure 6 . Download figure Open in new tab Fig 6. Reported time period for AI model prediction for CVD AI model development method In AI, there exists two types of learning: machine (which involves “the capacity of systems to learn from problem-specific training data to automate the process of analytical model building and solve associated tasks” 55 p.1 ), and deep learning (which uses artificial neural networks to emulate the human brain and its process of learning). 55 An overwhelming majority of studies in this review developed machine learning models (n=19) rather than deep learning models 31 , 35 , 41 , 42 , 44 , 46 , as shown in Figure 7 . Download figure Open in new tab Fig 7. The form of AI learning used in studies A variety of AI techniques were used for model development: random survival forest (n=12), proportional hazard technique (n=11) and artificial neural networks (n=7) 31 , 35 , 39 , 41 , 42 , 44 , 46 were the most common. Other techniques included gradient-boosting decision trees (n=6), 34 , 36 , 40 , 48 , 49 , 52 support vector machines (n=4) 29 , 30 , 34 , 40 and Naive Bayes classifiers (n=1). 43 The most common measure of model discrimination, or the process undertaken to identify the most accurate or useful AI model from a set of competing developed models, 56 was a form of an AUC (area under the curve) measure (n=18). Confidence intervals were also recorded in 10 studies. Specificity and sensitivity are measures which evaluated the number of true versus false negatives (specificity) or positives (sensitivity) that are reported by the AI model. 57 Sensitivity was a measure mentioned in seven included model evaluations, 33 , 34 , 35 , 38 , 42 , 43 , 46 while specificity measures were present in eight studies. 29 , 30 , 33 , 34 , 35 , 38 , 42 , 46 Calibration, or the comparison between predicted and actual results of patients, was only discussed in a minority of studies (n=6), 36 , 40 , 47 , 48 , 51 , 53 despite its importance in model development. Study limitations and model validation The most commonly mentioned limitations were limitations with the methodology or development of the model, which were mentioned in 17 papers. These were then followed by limitations in AI model validation (n=14), dataset limitations (n=11), as well as limitations with the AI or technological aspects of the models (n=6). 30 , 35 , 43 , 44 , 52 , 53 Specifically on model validation, it was found that the majority of models included had reached an internal validation state (n=18, 72%). While a few papers were able to conduct external validations (n=5), 30 , 34 , 36 , 48 , 49 there were no cases in which a clinical trial was conducted to evaluate the effectiveness of these models, as shown in Figure 8 . Download figure Open in new tab Fig 8. Graph of the AI model stage of development reached by studies in this review Discussion Since 2021, the environment surrounding AI risk models has grown substantially and more accessible to the average researcher or clinician due to the increasing presence of AI practices in daily procedures. 58 Overall, the aim was to 1) assess the current stage of development these models have achieved, 2) to identify the AI techniques used in the creation and validation of the models, and 3) to pinpoint potential sources of bias or limitations in the identified AI models. Findings of this study identified the overall trends with model development, specifically the data and algorithmic biases present in the models included, as described below. Major data bias findings Data biases, or bias resulting from the presence of unfair or unrepresentative data that helped develop the AI model, 59 may lead to inaccurate CVD predictions or biased predictions associated with certain population groups. 60 Geographic discrepancy Through the population analysis of the included studies, an overwhelming majority of databases were based in the United States of America or in Eastern Asia (i.e. China, Japan, Taiwan, Korea, etc.). However, there were no datasets used from the African or Oceania or Middle Eastern regions, highlighting a significant gap in the knowledge. This finding has been mentioned in previous reviews of CVD AI risk models. 21 , 50 Additionally, several studies did not provide statistics regarding the age, sex, or racial distribution of their study cohort, not fully explaining any underlying biases or over/under presence of certain populations. This discrepancy in data usage and provision from different geographic regions shows the importance of using multi centred studies should be encouraged. Consistency among CVD predictors and predictions Overall, age was found to be the most commonly used predictor of CVD or related events, having been mentioned in models from 18 papers. BMI, having a history of smoking/or being a current smoker, as well as a history of hypertension and the patient’s sex were also found to be parameters that were significant in predictions, due to their overwhelming use in nine or more studies. Standardisation, or the technical specifications necessary for AI (and some other technologies) to perform effectively, was highlighted by Golenkov et al 61 as a novel difficulty for AI in its staggering process. This is as standardisation (or lack thereof) may impact the compatibility and interoperability between models or different standards. 50 Creating a “gold standard”, or a commonly agreed upon method of measurement, or measure where a person is categorised as having developed a disease is important, as it allows for two AI models to be used and predict a similar outcome, even if they use different types of learning or parameters. 60 Similarly, the most common outcome prediction of the models was an unspecified CVD diagnosis category. While specific cardiovascular diseases, such as heart failure or atrial fibrillation are acceptable outcomes for these models, CVD, which is a “collective term designating all types of affliction affecting the blood circulatory system, including the heart and vasculature”, 62 p.1 is not specific enough to be able to understand what exactly is being predicted by the model. Major algorithmic bias findings On the other hand, algorithmic biases, or biases introduced by the AI algorithms themselves, may also lead to inaccurate CVD predictions or the impression of AI models to make predictions treating groups fairly. 63 Model development and its impact on clinical practice The results of this study found a lack of development of AI risk prediction CVD models, especially in relation to independent external validation. The goal of the majority of the included studies is to aid in the individualised diagnosis or risk calculation of a patient in regards to CVD, and this can only be properly achieved on a worldwide basis if there is a focus on (specifically external) model validation. 64 This overall trend of an large number of internally validated sources may be due to a focus by researchers on the development of new and diverse models, rather than the testing of previously designed risk predictor models in environments or with patients previously unseen or unused during the models derivation or internal validation, such as validation the model among a population of a different background or ethnicity. According to the six-stage process as outlined by de Hond et al. (2022) 50 , 20 of the included models in this review have only reached stage two: Development of the AI prediction model (which includes both derivation and internal validation). 50 It is also important to note that no model surpasses stage three: Validation of the AI prediction model, as no study was found to have undergone clinical trial to not only test for the accuracy of the model, but also its usefulness, such as with patient or doctor-AI interactions, addressing AI security risks, and impact assessment of the model in clinical practice. 50 Model discrimination, which are the biases and discriminatory outcomes that arise when AI risk prediction models make predictions or decisions, 56 was found to have been investigated in 92% of included studies. Discrimination in model performance or predictions made, whether explainable or unexplainable, may arise from bias in the data and/or the algorithm, 65 and may have both legal and ethical considerations, such as the neglect of certain ethical groups or geographic locations. 21 Model calibration measures is the process of adjusting the predicted probabilities of a model so that they accurately represent the true likelihood of an event, 66 was conducted in only 24% of models. Conducting these calibration tests is crucial to providing accurate risk predictions to patients, as to avoid providing incorrect real-life medical interventions to patients. 26 , 67 Due to the emerging environment around AI in healthcare decision-making, it is crucial that these models undergo both discrimination and calibration checks, as to ensure these systems are both fair and trustworthy. 68 , 69 Implications of AI risk prediction models for CVD in PC or AC This research underscores the critical role of AI in improving the prediction and prevention of cardiovascular diseases, while also highlighting the existing gaps and challenges in the application and validation of these models in primary and ambulatory care. A focus regarding indemnifying gaps in the pre-existing literature, specifically regarding the presence of algorithmic and data biases, primarily on the importance of external validation and clinical trials, as in stark contrast to the current trend of prioritizing only the development and internal validation of these models. Clinical researchers may be able to externally validate models, such as ones included in this review or newly created models, as to aid in discovering a model that is accurate as well as clinically relevant and useful for primary or ambulatory care clinicians. This also applies to global research funding bodies, which should be focusing on funding projects taking the step to externally validate new or existing models, as well as creating a standardisation of models and their parameters, and/or global datasets used, rather than solely promoting the development of novel AI models following in the steps of previous study methodology. Strengths & limitations Study strengths The use of a scoping review methodology was a significant strength of this article, due to the rigorous screening and assessment of included studies, which was conducted using the Arksey and O’Malley scoping review framework. 24 This study was rigorous in its adherence to the method, ensuring all steps from the development of the research question to the consultation of the findings with other experts in the field were thoroughly followed. Study limitations There are several limitations to this review, including a limit to only including full text, accessible, English language studies. This may have led to an incomplete scope of the existing literature or models that have been researched related to this review, and may have impacted the findings of this study, such as the global distribution of datasets used. As AI risk prediction models is a rapidly evolving field, conducting this review even a month later, may have also produced different results, such as additional models. As well, this review did not aim to summarize the clinical applicability or effectiveness of the models, especially compared to traditional or non AI risk prediction models. While it is important to understand the general structure and key aspects of these models (such a time to events or parameters included), their usefulness is a very important investigation, as to continue to improve upon the models, and the future of AI in healthcare in general. This makes the comparison between AI and non-AI models an essential area for further research. Within the 25 included studies, none used the precision recall AUC, which is a useful tool in model discrimination, in which there is a focus on achieving a high number of true positives (where the model was correct in its prediction for a given patient). AUC-ROC tends to be a more “simplistic” evaluation of a risk prediction model (though still useful in the validation of the model), whereas a precision recall AUC may be more beneficial in cases where the true-positive and true–negative outcome are not balanced in their costs (development of CVD or not). 70 This highlights the need to define measures of model accuracy that can be used across studies. Finally, in the scoping review methodology, there is no formal evaluation of the quality of the included studies as is typical in systematic reviews. Conclusion In conclusion, there are a significant number of studies involved in the development and internal validation of novel and diverse AI risk prediction models for CVD in a PC or AC environment. This scoping review highlights that there needs to be a multi centred, standardisation, and external validation focus to these models going forward, as to continue to discover the clinical utility of risk prediction models using AI. Gaps identified by authors will act as future guidance for PC and AC researchers, research funding avenues, and governmental policies regarding AI usage and development. Data Availability All data produced in the present work are contained in the manuscript Acknowledgements Thank you to the University College Dublin School of Medicine for generously funding this research through the AI in Medicine Scholarship and for facilitating its development through the Student Summer Research Award (SSRA) program. Appendix 1 (Heart failure [MeSH Terms] OR Atrial fibrillation [MeSH Terms] OR Acute coronary syndrome [MeSH Terms] OR Cardiovascular diseases [MeSH Terms] OR Myocardial infarction [MeSH Terms] OR coronary Artery disease [MeSH Terms] OR stroke [MeSH Terms] OR cerebrovascular disorders [MeSH Terms] OR heart disease [MeSH Terms] OR arrhythmias, Cardiac [MeSH Terms] OR Angina, Pectoris [MeSH Terms] OR myocardial ischemia [MeSH Terms] OR All-cause mortality [MeSH Terms] OR Coronary artery bypass [MeSH Terms] OR Percutaneous transluminal coronary angioplasty [MeSH Terms] OR Heart failure [Text Word] OR Atrial fibrillation [Text Word] OR Acute coronary syndrome [Text Word] OR Cardiovascular diseases [Text Word] OR Myocardial infarction[Text Word] OR coronary Artery disease [Text Word] OR stroke [Text Word] OR cerebrovascular disorders [Text Word] OR heart disease [Text Word] OR arrhythmias, Cardiac [Text Word] OR Angina, Pectoris [Text Word] OR myocardial ischemia [Text Word] OR All-cause mortality [Text Word] OR Coronary artery bypass [Text Word] OR Percutaneous transluminal coronary angioplasty [Text Word]) AND (Machine learning [MeSH Terms] OR Deep learning [MeSH Terms] OR Artificial intelligence [MeSH Terms] OR Unsupervised [Text Word] OR Supervised [Text Word] OR Deep learning [Text word] OR Machine learning [Text Word] OR Artificial intelligence [Text Word]) AND (Risk [MeSH Terms] OR Mortality [MeSH Terms] OR Risk prediction [Title/Abstract] OR Prediction [Text Word]) Appendix 2 View this table: View inline View popup Download powerpoint References 1. ↵ Roth GA , Mensah GA , Johnson CO , Addolorato G , Ammirati E , Baddour LM , et al. Global burden of cardiovascular diseases and risk factors, 1990–2019 . Journal of the American College of Cardiology . 2020 Dec 22; 76 ( 25 ): 2982 – 3021 . doi: 10.1016/j.jacc.2020.11.010 OpenUrl CrossRef PubMed 2. ↵ Nichols M , Townsend N , Scarborough P , Rayner M . European Cardiovascular Disease Statistics . Brussels: European Heart Network:LJ : European Society of Cardiology ; 2012 . 3. ↵ Janes H , Pepe MS , Gu W . Assessing the value of risk predictions by using risk stratification tables . Annals of Internal Medicine . 2008 Nov 18; 149 ( 10 ): 751 – 60 . doi: 10.7326/0003-4819-149-10-200811180-00009 OpenUrl CrossRef PubMed Web of Science 4. ↵ Bomrah S , Uddin M , Upadhyay U , Komorowski M , Priya J , Dhar E , et al. A scoping review of machine learning for sepsis prediction-feature engineering strategies and model performance: A step towards explainability . Critical Care . 2024 May 28; 28 ( 1 ). doi: 10.1186/s13054-024-04948-6 OpenUrl CrossRef 5. ↵ Sun H , Depraetere K , Meesseman L , Cabanillas Silva P , Szymanowsky R , Fliegenschmidt J , et al. Machine learning–based prediction models for different clinical risks in different hospitals: Evaluation of Live Performance . Journal of Medical Internet Research . 2022 Jun 7; 24 ( 6 ). doi: 10.2196/34295 OpenUrl CrossRef 6. ↵ Lloyd-Jones DM . Cardiovascular risk prediction . Circulation . 2010 Apr 20; 121 ( 15 ): 1768 – 77 . doi: 10.1161/circulationaha.109.849166 OpenUrl FREE Full Text 7. ↵ Collins GS , Dhiman P , Ma J , Schlussel MM , Archer L , Van Calster B , et al. Evaluation of clinical prediction models (part 1): From development to external validation . BMJ . 2024 Jan 8; 384 ( e074819 ). doi: 10.1136/bmj-2023-074819 OpenUrl FREE Full Text 8. ↵ Riley RD , Pate A , Dhiman P , Archer L , Martin GP , Collins GS . Clinical prediction Models and the Multiverse of Madness . BMC Medicine . 2023 Dec 18; 21 ( 1 ). doi: 10.1186/s12916-023-03212-y OpenUrl CrossRef PubMed 9. ↵ Hassan N , Slight R , Morgan G , Bates DW , Gallier S , Sapey E , et al. Road map for clinicians to develop and evaluate AI predictive models to inform clinical decision-making . BMJ Health Care Inform . 2023 Aug ; 30 ( 1 ). doi: 10.1136/bmjhci-2023-100784 OpenUrl Abstract / FREE Full Text 10. ↵ Kuziemski M , Misuraca G . AI governance in the Public Sector: Three tales from the frontiers of Automated Decision-making in Democratic settings . Telecommunications Policy . 2020 Jul 17; 44 ( 6 ): 101976 . doi: 10.1016/j.telpol.2020.101976 OpenUrl CrossRef 11. ↵ Banerjee A , Chen S , Fatemifar G , Zeina M , Lumbers RT , Mielke J , et al. Machine learning for subtype definition and risk prediction in heart failure, acute coronary syndromes and atrial fibrillation: Systematic review of validity and clinical utility . BMC Medicine . 2021 Apr 6; 19 ( 1 ). doi: 10.1186/s12916-021-01940-7 OpenUrl CrossRef 12. ↵ Khalifa M , Albadawy M . Artificial Intelligence for clinical prediction: Exploring key domains and essential functions . Computer Methods and Programs in Biomedicine Update . 2024 ; 5 : 100148 . doi: 10.1016/j.cmpbup.2024.100148 OpenUrl CrossRef 13. ↵ Lee S , Chu Y , Ryu J , Park YJ , Yang S , Koh SB . Artificial Intelligence for detection of cardiovascular-related diseases from wearable devices: A systematic review and meta-analysis . Yonsei Medical Journal . 2022 Jan 6; 63 ( Suppl ). doi: 10.3349/ymj.2022.63.s93 OpenUrl CrossRef 14. ↵ Triantafyllidis A , Kondylakis H , Katehakis D , Kouroubali A , Koumakis L , Marias K , et al. Deep learning in mhealth for cardiovascular disease, diabetes, and cancer: Systematic review . JMIR mHealth and uHealth . 2022 Apr 4; 10 ( 4 ). doi: 10.2196/32344 OpenUrl CrossRef 15. ↵ Kee OT , Harun H , Mustafa N , Abdul Murad NA , Chin SF , Jaafar R , et al. Cardiovascular complications in a diabetes prediction model using Machine Learning: A Systematic Review . Cardiovascular Diabetology . 2023 Jan 19; 22 ( 1 ). doi: 10.1186/s12933-023-01741-7 OpenUrl CrossRef PubMed 16. ↵ Andaur Navarro CL , Damen JA , Takada T , Nijman SW , Dhiman P , Ma J , et al. Risk of bias in studies on prediction models developed using supervised machine learning techniques: Systematic review . BMJ . 2021 Oct 20; doi: 10.1136/bmj.n2281 OpenUrl Abstract / FREE Full Text 17. ↵ Unceta I , Nin J , Pujol O . Risk mitigation in algorithmic accountability: The role of machine learning copies . PLOS ONE . 2020 Nov 3; 15 ( 11 ). doi: 10.1371/journal.pone.0241286 OpenUrl CrossRef 18. ↵ Fehr J , Citro B , Malpani R , Lippert C , Madai VI . A trustworthy AI reality-check: The lack of transparency of artificial intelligence products in Healthcare . Frontiers in Digital Health . 2024 Feb 20; 6 . doi: 10.3389/fdgth.2024.1267290 OpenUrl CrossRef 19. ↵ Liu R , Wang M , Zheng T , Zhang R , Li N , Chen Z , et al. An artificial intelligence-based risk prediction model of myocardial infarction . BMC Bioinformatics . 2022 Jun 7; 23 ( 1 ). doi: 10.1186/s12859-022-04761-4 OpenUrl CrossRef 20. ↵ Hughes JW , Tooley J , Torres Soto J , Ostropolets A , Poterucha T , Christensen MK , et al. A deep learning-based electrocardiogram risk score for long term cardiovascular death and disease . NPJ Digital Medicine . 2023 Sept 12 ; 6 ( 1 ). doi: 10.1038/s41746-023-00916-6 OpenUrl CrossRef PubMed 21. ↵ Cai Y , Cai Y-Q , Tang L-Y , Wang Y-H , Gong M , Jing T-C , et al. Artificial Intelligence in the risk prediction models of cardiovascular disease and development of an independent Validation Screening Tool: A systematic review . BMC Medicine . 2024 Feb 5; 22 ( 1 ). doi: 10.1186/s12916-024-03273-7 OpenUrl CrossRef 22. ↵ Silva GF , Fagundes TP , Teixeira BC , Chiavegatto Filho AD . Machine Learning for Hypertension Prediction: A systematic review . Current Hypertension Reports . 2022 Jun 22; 24 ( 11 ): 523 – 33 . doi: 10.1007/s11906-022-01212-6 OpenUrl CrossRef PubMed 23. ↵ Thomas A , Lubarsky S , Durning SJ , Young ME . Knowledge syntheses in medical education: Demystifying scoping reviews . Academic Medicine . 2017 Feb ; 92 ( 2 ): 161 – 6 . doi: 10.1097/acm.0000000000001452 OpenUrl CrossRef PubMed 24. ↵ Arksey H , O’Malley L . Scoping studies: Towards a methodological framework . International Journal of Social Research Methodology . 2005 Feb ; 8 ( 1 ): 19 – 32 . doi: 10.1080/1364557032000119616 OpenUrl CrossRef 25. ↵ Levac D , Colquhoun H , O’Brien KK . Scoping studies: Advancing the methodology . Implementation Science . 2010 Sept 20; 5 ( 1 ). doi: 10.1186/1748-5908-5-69 OpenUrl CrossRef PubMed 26. ↵ Collins GS , Moons KG , Dhiman P , Riley RD , Beam AL , Van Calster B , et al. TRIPOD+AI statement: Updated guidance for reporting clinical prediction models that use regression or machine learning methods . BMJ . 2024 Apr 16; 385 . doi: 10.1136/bmj-2023-078378 OpenUrl FREE Full Text 27. ↵ Khera AV , Kathiresan S . Genetics of coronary artery disease: Discovery, Biology and clinical translation . Nature Reviews Genetics . 2017 Mar 13; 18 ( 6 ): 331 – 44 . doi: 10.1038/nrg.2016.160 OpenUrl CrossRef PubMed 28. ↵ Friedman J , Hastie T , Tibshirani R. The elements of statistical learning data mining, Inference, and prediction . New York, NY : Springer-Verlag New York ; 2009 . 29. ↵ Qian X , Li Y , Zhang X , Guo H , He J , Wang X , et al. A cardiovascular disease prediction model based on routine physical examination indicators using machine learning methods: A cohort study . Frontiers in Cardiovascular Medicine . 2022 Jun 17; 9 . doi: 10.3389/fcvm.2022.854287 OpenUrl CrossRef 30. ↵ Kao Y-T , Huang C-Y , Fang Y-A , Liu J-C , Chang T-H . Machine learning-based prediction of atrial fibrillation risk using electronic medical records in older aged patients . The American Journal of Cardiology . 2023 May ; 198 : 56 – 63 . doi: 10.1016/j.amjcard.2023.03.035 OpenUrl CrossRef PubMed 31. ↵ Butler L , Karabayir I , Kitzman DW , Alonso A , Tison GH , Chen LY , et al. A generalizable electrocardiogram-based artificial intelligence model for 10-year heart failure risk prediction . Cardiovascular Digital Health Journal . 2023 Dec ; 4 ( 6 ): 183 – 90 . doi: 10.1016/j.cvdhj.2023.11.003 OpenUrl CrossRef PubMed 32. ↵ Greenberg B , Adler E , Campagnari C , Yagil A . A machine learning risk score predicts mortality across the spectrum of left ventricular ejection fraction . European Journal of Heart Failure . 2021 Apr 6; 23 ( 6 ): 995 – 9 . doi: 10.1002/ejhf.2155 OpenUrl CrossRef PubMed 33. ↵ Grout RW , Hui SL , Imler TD , El-Azab S , Baker J , Sands GH , et al. Development, validation, and proof-of-concept implementation of a two-year risk prediction model for undiagnosed atrial fibrillation using common electronic health data (UNAFIED) . BMC Medical Informatics and Decision Making . 2021 Apr 3; 21 ( 1 ). doi: 10.1186/s12911-021-01482-1 OpenUrl CrossRef 34. ↵ Petrazzini BO , Chaudhary K , Márquez-Luna C , Forrest IS , Rocheleau G , Cho J , et al. Coronary risk estimation based on clinical data in Electronic Health Records . Journal of the American College of Cardiology . 2022 Mar 29; 79 ( 12 ): 1155 – 66 . doi: 10.1016/j.jacc.2022.01.021 OpenUrl CrossRef PubMed 35. ↵ Raghunath S , Pfeifer JM , Ulloa-Cerna AE , Nemani A , Carbonati T , Jing L , et al. Deep neural networks can predict new-onset atrial fibrillation from the 12-lead ECG and help identify those at risk of atrial fibrillation–related stroke . Circulation . 2021 Mar 30; 143 ( 13 ): 1287 – 98 . doi: 10.1161/circulationaha.120.047829 OpenUrl CrossRef PubMed 36. ↵ Segar MW , Jaeger BC , Patel KV , Nambi V , Ndumele CE , Correa A , et al. Development and validation of machine learning–based race-specific models to predict 10-year risk of heart failure: A multicohort analysis . Circulation . 2021 Jun 15; 143 ( 24 ): 2370 – 83 . doi: 10.1161/circulationaha.120.053134 OpenUrl CrossRef PubMed 37. ↵ Tison GH , Avram R , Nah G , Klein L , Howard BV , Allison MA , et al. Predicting incident heart failure in women with Machine Learning: The women’s health initiative cohort . Canadian Journal of Cardiology . 2021 Nov ; 37 ( 11 ): 1708 – 14 . doi: 10.1016/j.cjca.2021.08.006 OpenUrl CrossRef PubMed 38. ↵ Wang C , Zhao Y , Jin B , Gan X , Liang B , Xiang Y , et al. Development and validation of a predictive model for coronary artery disease using machine learning . Frontiers in Cardiovascular Medicine . 2021 Feb 2; 8 . doi: 10.3389/fcvm.2021.614204 OpenUrl CrossRef 39. ↵ Al-Droubi SS , Jahangir E , Kochendorfer KM , Krive M , Laufer-Perl M , Gilon D , et al. Artificial intelligence modelling to assess the risk of cardiovascular disease in oncology patients . European Heart Journal - Digital Health . 2023 May 8; 4 ( 4 ): 302 – 15 . doi: 10.1093/ehjdh/ztad031 OpenUrl CrossRef PubMed 40. ↵ Wang Y , Hou R , Ni B , Jiang Y , Zhang Y . Development and validation of a prediction model based on machine learning algorithms for predicting the risk of heart failure in middle:LJaged and older US people with Prediabetes or diabetes . Clinical Cardiology . 2023 Jul 31; 46 ( 10 ): 1234 – 43 . doi: 10.1002/clc.24104 OpenUrl CrossRef PubMed 41. ↵ Morris MC , Moradi H , Aslani M , Sims M , Schlundt D , Kouros CD , et al. Predicting incident cardiovascular disease among African-American adults: A deep learning approach to evaluate social determinants of health in the Jackson Heart Study . PLOS ONE . 2023 Nov 10; 18 ( 11 ). doi: 10.1371/journal.pone.0294050 OpenUrl CrossRef 42. ↵ Habineza T , Ribeiro AH , Gedon D , Behar JA , Ribeiro AL , Schön TB . End-to-end risk prediction of atrial fibrillation from the 12-lead ECG by Deep Neural Networks . Journal of Electrocardiology . 2023 ; 81 : 193 – 200 . doi: 10.1016/j.jelectrocard.2023.09.011 OpenUrl CrossRef PubMed 43. ↵ Harada D , Asanoi H , Noto T , Takagawa J . Naive Bayes prediction of the development of cardiac events in heart failure with preserved ejection fraction in an outpatient clinic - beyond B-type natriuretic peptide . Circulation Journal . 2021 Dec 24; 86 ( 1 ): 37 – 46 . doi: 10.1253/circj.cj-21-0131 OpenUrl CrossRef PubMed 44. ↵ Baek Y-S , Lee D-H , Jo Y , Lee S-C , Choi W , Kim D-H . Artificial intelligence-estimated biological heart age using a 12-lead electrocardiogram predicts mortality and cardiovascular outcomes . Frontiers in Cardiovascular Medicine . 2023 Apr 13; 10 . doi: 10.3389/fcvm.2023.1137892 OpenUrl CrossRef PubMed 45. ↵ Heitzinger G , Spinka G , Koschatko S , Baumgartner C , Dannenberg V , Halavina K , et al. A streamlined, machine learning-derived approach to risk-stratification in heart failure patients with secondary tricuspid regurgitation . European Heart Journal - Cardiovascular Imaging . 2023 Feb 10; 24 ( 5 ): 588 – 97 . doi: 10.1093/ehjci/jead009 OpenUrl CrossRef PubMed 46. ↵ Heitzinger G , Spinka G , Koschatko S , Baumgartner C , Dannenberg V , Halavina K , et al. A streamlined, machine learning-derived approach to risk-stratification in heart failure patients with secondary tricuspid regurgitation . European Heart Journal - Cardiovascular Imaging . 2023 Feb 10; 24 ( 5 ): 588 – 97 . doi: 10.1093/ehjci/jead009 OpenUrl CrossRef PubMed 47. ↵ Morgenstern JD , Rosella LC , Costa AP , Anderson LN . Development of machine learning prediction models to explore nutrients predictive of cardiovascular disease using Canadian linked population-based data. Applied Physiology , Nutrition, and Metabolism . 2022 May ; 47 ( 5 ): 529 – 46 . doi: 10.1139/apnm-2021-0502 OpenUrl CrossRef 48. ↵ Deb B , Scott C , Pislaru SV , Nkomo VT , Kane GC , Alkhouli M , et al. Machine learning facilitates the prediction of long-term mortality in patients with tricuspid regurgitation . Open Heart . 2023 Nov 27; 10 ( 2 ). doi: 10.1136/openhrt-2023-002417 OpenUrl Abstract / FREE Full Text 49. ↵ Kobayashi M , Huttin O , Magnusson M , Ferreira JP , Bozec E , Huby A-C , et al. Machine learning-derived echocardiographic phenotypes predict heart failure incidence in asymptomatic individuals . JACC: Cardiovascular Imaging . 2022 Feb ; 15 ( 2 ): 193 – 208 . doi: 10.1016/j.jcmg.2021.07.004 OpenUrl CrossRef 50. ↵ de Hond AA , Leeuwenberg AM , Hooft L , Kant IM , Nijman SW , van Os HJ , et al. Guidelines and quality criteria for artificial intelligence-based prediction models in Healthcare: A scoping review . NPJ Digital Medicine . 2022 Jan 10; 5 ( 1 ). doi: 10.1038/s41746-021-00549-7 OpenUrl CrossRef PubMed 51. ↵ Ishii M , Kaikita K , Yasuda S , Akao M , Ako J , Matoba T , et al. Risk prediction score for clinical outcome in atrial fibrillation and stable coronary artery disease . Open Heart . 2023 May ; 10 ( 1 ). doi: 10.1136/openhrt-2023-002292 OpenUrl Abstract / FREE Full Text 52. ↵ Lu J , Bisson A , Bennamoun M , Zheng Y , Sanfilippo FM , Hung J , et al. Predicting multifaceted risks using machine learning in atrial fibrillation: Insights from Gloria-AF Study . European Heart Journal - Digital Health . 2024 Feb 20; 5 ( 3 ): 235 – 46 . doi: 10.1093/ehjdh/ztae010 OpenUrl CrossRef PubMed 53. ↵ Sun Q , Jiang S , Wang X , Zhang J , Li Y , Tian J , et al. A prediction model for major adverse cardiovascular events in patients with heart failure based on high-throughput echocardiographic data . Frontiers in Cardiovascular Medicine . 2022 Oct 28; 9 . doi: 10.3389/fcvm.2022.1022658 OpenUrl CrossRef 54. ↵ Anand V , Hu H , Weston AD , Scott CG , Michelena HI , Pislaru SV , et al. Machine learning-based risk stratification for mortality in patients with severe aortic regurgitation . European Heart Journal - Digital Health . 2023 Feb 7; 4 ( 3 ): 188 – 95 . doi: 10.1093/ehjdh/ztad006 OpenUrl CrossRef PubMed 55. ↵ Janiesch C , Zschech P , Heinrich K . Machine learning and deep learning . Electronic Markets . 2021 Apr 8; 31 ( 3 ): 685 – 95 . doi: 10.1007/s12525-021-00475-2 OpenUrl CrossRef 56. ↵ Bhat G , Danelljan M , Van Gool L , Timofte R . Learning discriminative model prediction for tracking . 2019 IEEE/CVF International Conference on Computer Vision (ICCV) . 2019 Oct; 6181 – 90 . doi: 10.1109/iccv.2019.00628 57. ↵ Wang H , Wang B , Zhang X , Feng C . Relations among sensitivity, specificity and predictive values of medical tests based on biomarkers . General Psychiatry . 2021 Mar 10; 34 ( 2 ). doi: 10.1136/gpsych-2020-100453 OpenUrl Abstract / FREE Full Text 58. ↵ Lattie EG , Stiles-Shields C , Graham AK . An overview of and recommendations for more accessible Digital Mental Health Services . Nature Reviews Psychology . 2022 Jan 26; 1 ( 2 ): 87 – 100 . doi: 10.1038/s44159-021-00003-1 OpenUrl CrossRef PubMed 59. ↵ Dissanayake H , Bracewell PJ . Bias in data-informed decision making . Encyclopedia of Data Science and Machine Learning . 2022 Oct 14; 1173 – 86 . doi: 10.4018/978-1-7998-9220-5.ch068 OpenUrl CrossRef 60. ↵ Estiri H , Strasser ZH , Rashidian S , Klann JG , Wagholikar KB , McCoy TH , et al. An objective framework for evaluating unrecognized bias in medical AI models predicting COVID-19 outcomes . Journal of the American Medical Informatics Association . 2022 May 12; 29 ( 8 ): 1334 – 41 . doi: 10.1093/jamia/ocac070 OpenUrl CrossRef PubMed 61. ↵ Golenkov V , Guliakina N , Golovko V , Krasnoproshin V . Artificial Intelligence Standardization is a key challenge for the technologies of the future . Communications in Computer and Information Science . 2020 Oct ; 1 – 21 . doi: 10.1007/978-3-030-60447-9_1 OpenUrl CrossRef 62. ↵ Thiriet M . Cardiovascular disease: An introduction . Biomathematical and Biomechanical Modeling of the Circulatory and Ventilatory Systems . 2019 Feb 19; 1 – 90 . doi: 10.1007/978-3-319-89315-0_1 OpenUrl CrossRef 63. ↵ Kordzadeh N , Ghasemaghaei M . Algorithmic bias: Review, synthesis, and future research directions . European Journal of Information Systems . 2021 Jun 6; 31 ( 3 ): 388 – 409 . doi: 10.1080/0960085x.2021.1927212 OpenUrl CrossRef 64. ↵ Ramspek CL , Jager KJ , Dekker FW , Zoccali C , van Diepen M . External validation of prognostic models: What, why, how, when and where? Clinical Kidney Journal . 2020 Nov 24; 14 ( 1 ): 49 – 58 . doi: 10.1093/ckj/sfaa188 OpenUrl CrossRef PubMed 65. ↵ Nazer LH , Zatarah R , Waldrip S , Ke JX , Moukheiber M , Khanna AK , et al. Bias in artificial intelligence algorithms and recommendations for mitigation . PLOS Digital Health . 2023 Jun 22; 2 ( 6 ). doi: 10.1371/journal.pdig.0000278 OpenUrl CrossRef PubMed 66. ↵ Wang C . Calibration in deep learning: A survey of the state-of-the-art . arXiv preprint arXiv:2308.01222. 2023 Aug 2 67. ↵ Hassan N , Slight R , Morgan G , Bates DW , Gallier S , Sapey E , et al. Road map for clinicians to develop and evaluate AI predictive models to inform clinical decision-making . BMJ Health Care Inform . 2023 Aug ; 30 ( 1 ). doi: 10.1136/bmjhci-2023-100784 OpenUrl Abstract / FREE Full Text 68. ↵ Varona D , Suárez JL . Discrimination, bias, fairness, and trustworthy AI . Applied Sciences . 2022 Jun 8; 12 ( 12 ): 5826 . doi: 10.3390/app12125826 OpenUrl CrossRef 69. ↵ Belenguer L . Ai Bias: Exploring discriminatory algorithmic decision-making models and the application of possible machine-centric solutions adapted from the pharmaceutical industry . AI and Ethics . 2022 Feb 10; 2 ( 4 ): 771 – 87 . doi: 10.1007/s43681-022-00138-8 OpenUrl CrossRef 70. ↵ Sofaer HR , Hoeting JA , Jarnevich CS . The area under the precision:LJrecall curve as a performance metric for rare binary events . Methods in Ecology and Evolution . 2019 Feb 14; 10 ( 4 ): 565 – 77 . doi: 10.1111/2041-210x.13140 OpenUrl CrossRef View the discussion thread. Back to top Previous Next Posted March 21, 2025. Download PDF Data/Code Email Thank you for your interest in spreading the word about medRxiv. NOTE: Your email address is requested solely to identify you as the sender of this article. Your Email * Your Name * Send To * Enter multiple addresses on separate lines or separate them with commas. You are going to email the following Artificial Intelligence (AI) Models for Cardiovascular Disease Risk Prediction in Primary and Ambulatory Care: A Scoping Review Message Subject (Your Name) has forwarded a page to you from medRxiv Message Body (Your Name) thought you would like to see this page from the medRxiv website. Your Personal Message CAPTCHA This question is for testing whether or not you are a human visitor and to prevent automated spam submissions. Share Artificial Intelligence (AI) Models for Cardiovascular Disease Risk Prediction in Primary and Ambulatory Care: A Scoping Review Caoimhe Provost , John Broughan , Geoff McCombe , Mark O Kelly , Mark Ledwidge , Walter Cullen , Joseph Gallagher medRxiv 2025.03.21.25324379; doi: https://doi.org/10.1101/2025.03.21.25324379 Share This Article: Copy Citation Tools Artificial Intelligence (AI) Models for Cardiovascular Disease Risk Prediction in Primary and Ambulatory Care: A Scoping Review Caoimhe Provost , John Broughan , Geoff McCombe , Mark O Kelly , Mark Ledwidge , Walter Cullen , Joseph Gallagher medRxiv 2025.03.21.25324379; doi: https://doi.org/10.1101/2025.03.21.25324379 Citation Manager Formats BibTeX Bookends EasyBib EndNote (tagged) EndNote 8 (xml) Medlars Mendeley Papers RefWorks Tagged Ref Manager RIS Zotero Tweet Widget Facebook Like Google Plus One Subject Area Cardiovascular Medicine Subject Areas All Articles Addiction Medicine (569) Allergy and Immunology (863) Anesthesia (300) Cardiovascular Medicine (4442) Dentistry and Oral Medicine (444) Dermatology (383) Emergency Medicine (609) Endocrinology (including Diabetes Mellitus and Metabolic Disease) (1510) Epidemiology (15230) Forensic Medicine (30) Gastroenterology (1126) Genetic and Genomic Medicine (6608) Geriatric Medicine (668) Health Economics (998) Health Informatics (4542) Health Policy (1370) Health Systems and Quality Improvement (1613) Hematology (543) HIV/AIDS (1266) Infectious Diseases (except HIV/AIDS) (15923) Intensive Care and Critical Care Medicine (1103) Medical Education (623) Medical Ethics (147) Nephrology (668) Neurology (6607) Nursing (346) Nutrition (999) Obstetrics and Gynecology (1146) Occupational and Environmental Health (957) Oncology (3336) Ophthalmology (974) Orthopedics (369) Otolaryngology (420) Pain Medicine (436) Palliative Medicine (130) Pathology (664) Pediatrics (1693) Pharmacology and Therapeutics (692) Primary Care Research (712) Psychiatry and Clinical Psychology (5448) Public and Global Health (9237) Radiology and Imaging (2201) Rehabilitation Medicine and Physical Therapy (1370) Respiratory Medicine (1196) Rheumatology (596) Sexual and Reproductive Health (714) Sports Medicine (530) Surgery (712) Toxicology (99) Transplantation (289) Urology (265) (function(){function c(){var b=a.contentDocument||a.contentWindow.document;if(b){var d=b.createElement('script');d.innerHTML="window.__CF$cv$params={r:'a018878cfc061b23',t:'MTc3OTc1MzkzOQ=='};var a=document.createElement('script');a.src='/cdn-cgi/challenge-platform/scripts/jsd/main.js';document.getElementsByTagName('head')[0].appendChild(a);";b.getElementsByTagName('head')[0].appendChild(d)}}if(document.body){var a=document.createElement('iframe');a.height=1;a.width=1;a.style.position='absolute';a.style.top=0;a.style.left=0;a.style.border='none';a.style.visibility='hidden';document.body.appendChild(a);if('loading'!==document.readyState)c();else if(window.addEventListener)document.addEventListener('DOMContentLoaded',c);else{var e=document.onreadystatechange||function(){};document.onreadystatechange=function(b){e(b);'loading'!==document.readyState&&(document.onreadystatechange=e,c())}}}})();